JP5058985B2 - Point preselection for fast deformable point-based imaging - Google Patents

Description

  The present invention relates to a method and apparatus for point preselection for fast deformable point-based imaging.

  Radiotherapy treats diseases such as cancer tumors using radiation such as X-rays. In the process of applying radiation to the pathological tissue, some healthy tissue is also exposed to radiation. Exposure of healthy tissue to radiation can result in treatment-related complications. Therefore, it is desirable to accurately and accurately outline the pathological region so that most of the radiation is applied to the pathological tissue and minimally applied to the surrounding healthy tissue.

  Precisely and accurately delineating a treatment area (planned target volume, also known as PTV) incorporates the movement of that target during fractionated treatment. The movement can be a physical movement of the patient (setup error) or a movement and deformation of internal tissues including pathological tissues. These movements and deformations are caused by physical functions such as the heart system, respiratory system and digestive system or are obtained as a result of response to treatment. In conventional treatment planning, PTV is based on patient population statistics, which results in the target area being too large or inaccurate. In order to assess patient-specific motion, successive images are taken over a period of time to obtain a 3D depiction of the pathological tissue and surrounding organ geometry. Time sampling can be performed using 4D gated imaging, for example in seconds, monitoring respiratory motion, or daily and weekly, or a combination thereof. Examples of the combination include once a week imaging using a 4D gate imaging technique. Integrating temporal samples of 3D images in radiation therapy is commonly referred to as image-guided radiation therapy (IGRT) or adaptive radiation therapy. Such adjustment allows for more precise application of radiation to the target area.

  Quantitative measurement of the deformation process that occurs during treatment is necessary for many applications in IGRT. The dose volume histogram (DVH) is an important tool for assessing the amount of treatment that statistically describes the dose distribution in the target area and dangerous organs. DVH is traditionally based on a single 3D image. In order to accumulate a four-dimensional dose histogram, the spatial correspondence between the volume elements of the target area and the volume elements of the surrounding organs needs to be calculated in all images. This can be done by a deformable registration. Voxel-based registration methods are usually inefficient due to changes caused by movements and mass movements such as gas and excrement in the intestine. This is because it is very difficult to match the corresponding points in the image. Also, registration methods based on geometric correspondence (eg between points or surfaces) can be used. Model-based segmentation methods have been used to cleanly segment structures in a region of interest by automatically deforming a triangular surface mesh into the surface of the region of interest. If this technique is applied to all structures in all images, the apex of the adapted mesh defines a corresponding landmark between the reference object surface and the imaged object surface. The landmarks can be aligned and the deformation field for all voxels in the image can be estimated using a deformable point-based registration method, resulting in a volumetric deformation field that represents the motion of each voxel in the image. Give rise to

  The deformable model has the ability to adapt to the important deformability of biological structures. However, such a model is complex. Usually, the amount of vertices defined for a anatomical object is in units of a thousand. Calculating volumetric deformation fields using point-based registration techniques based on thousands of vertices or landmarks requires significant processing time, which is not feasible in a clinical setting. As such, there is a need for a deformation process that can provide an estimated volumetric deformation for a anatomical object within a clinically acceptable time period.

  The present invention is directed to a deformation model that produces volumetric image objects within a clinically acceptable time period. The deformation model systematically selects a smaller number of vertices that will be used in the deformable registration. The small number of vertices used allows for acceptable processing time. The systematic selection of vertices used provides that the selected vertices are distributed over the entire surface of the object being imaged, reducing the topographical error introduced by adjacent vertices.

  In one embodiment, a method is provided for reducing the number of corresponding vertex pairs based on a selected distance. The method sorts the corresponding vertex pairs in descending order based on the distance between the vertices of the vertex pair. The corresponding vertex pair with the greatest distance is given the highest priority. Corresponding vertex pairs that are within the selected distance are excluded, thereby reducing the number of corresponding vertex pairs used in the deformable registration. In some embodiments, the selected distance is the distance between the vertices of the selected corresponding vertex pair. In other embodiments, the selected distance is predetermined. In some embodiments, the distance can be varied.

  Embodiments of the present invention are illustrated in the corresponding drawings, which are incorporated in and constitute a part of this specification, together with the general description of the invention described above. The following detailed description serves as illustrative of the principles of the invention. Those skilled in the art will appreciate that these embodiments do not limit the invention, but merely provide examples that incorporate the principles of the invention.

  The deformation model disclosed herein provides for generating volumetric image objects within a clinically acceptable time period. The deformation model allows the reconstruction of an imaging object that indicates that it is moving. The deformation model makes it possible to more accurately describe the contour of the target of interest, thereby enabling more accurate radiation therapy. Although this disclosure discusses the applicability of the present invention to image-guided radiation therapy, one skilled in the art will understand that the disclosed deformation model is equally applicable to other imaging processes and modalities. It will be. In particular, the present invention can be applied to the case where the attention area is in motion over a time period.

  The disclosed deformation model provides a way to reduce the number of vertices used in the deformation. Thereby, it allows a reasonable processing time for clinical use. Reducing vertices must be such that it is possible to reduce processing time without reducing image accuracy. In this regard, the deformation model disclosed herein provides for reducing vertices in such a way that selected vertices are distributed across the surface of the object being imaged. The selected vertices represent all parts of the surface and are far enough away to substantially reduce the topological violations caused by the elastic point-based registration of closely located landmarks. Can be spaced.

  In one embodiment of the invention, the deformation model starts at step 10 in FIG. Pairs of triangular surface meshes of corresponding objects such as the bladder shown in FIGS. 1 and 2 are input at step 20. In step 30 in FIG. 4, the position of each vertex of the triangular surface mesh is recorded in the vertex index. Note that in many cases objects have the same number of vertices, but it is not necessary for two objects to have the same number of vertices for this method to be effective. In step 35, the individual vertices in the two triangular surface meshes are each matched, resulting in a corresponding set of vertex pairs. If two meshes have the same topology, the corresponding vertex pair is determined by the vertex index. If not, the corresponding vertex can be determined, for example, as closest in the sense of Euclidean distance, as closest along the surface normal at the vertex location, or using other methods.

  The Euclidean distance between each corresponding vertex pair is calculated in step 38 and the distance is stored in descending order in step 40. As such, the highest priority is given to the corresponding vertex pair with the greatest distance. The corresponding vertex pair with the maximum distance is selected at step 45. Thereafter, any corresponding vertex pair that is within the radius of the selected corresponding vertex pair is removed in step 48. This can best be understood with reference to FIG. FIG. 5 shows an overlay of a couple of corresponding vertex pairs, where the vertices from the first object are drawn as solid lines and the vertices from the second object are drawn as dotted lines. If points A and B are determined to be the corresponding vertex pair having the maximum distance, a radius R representing the distance between the selected corresponding vertex pairs is determined. Tracing within the area of radius R around point A shows that points C and D are within radius R while point E is outside radius R. Step 48 will remove points C and D and the corresponding vertices from the second imaging object and the corresponding vertex pair list. The removed points are not used in the deformable registration.

  Once the corresponding vertex pair that is within the radius of the selected vertex pair is excluded, the method proceeds to step 50 to determine if any additional corresponding vertex pairs remain in the vertex pair list. It is determined. If not, i.e., if all corresponding vertex pairs are either selected or excluded, the method proceeds to step 60 where deformable registration is performed. If the corresponding vertex pair remains in the corresponding vertex pair list, the method returns to step 45 to determine the next corresponding vertex pair having the maximum distance that has not been previously selected or excluded. To do. The method then proceeds to step 48 where the corresponding points within the radius of the newly selected corresponding vertex pair are excluded. This process is repeated until all corresponding vertex pairs are either selected or excluded.

  As a result of this process, the number of corresponding vertex pairs, usually in the thousands, is systematically reduced from about 1/10 to 1/100 and even more. For example, an anatomical object having thousands of corresponding vertex pairs can be reduced to 50 to 100 corresponding vertex pairs. The number of selected vertex pairs will vary depending on a number of factors including the size of the object and the amount of relative geometric change between the images. The low number of vertices allows a clinically acceptable processing time for deformable registration of the object being imaged. The method also provides for selecting vertices that are diffused across the surface of the entire object. This ensures that all parts of the object are represented by deformable registrations and prevents the use of adjacent vertex pairs that may result in topographical errors.

  In another embodiment of the deformable registration method, the corresponding vertex pair to be excluded is not determined by the radius of the selected corresponding vertex pair, but instead is based on a predetermined distance. In certain embodiments, the predetermined distance can be changed by a physician. In some embodiments, a step may be added that asks the physician to enter a distance. Thereafter, the method is performed to determine the number of selected corresponding vertex pairs. The method then asks the physician if the selected number of corresponding pairs is within a desired range prior to performing the deformable registration. If the number of selected corresponding vertex pairs is not within the desired range, the physician can change the predetermined distance and the method determines that the new number of selected corresponding vertex pairs is determined. enable. Also, the number of selected vertex pairs can have a threshold range. If the number of corresponding vertex pairs selected is within the threshold range, the deformable registration is performed. If the number of selected corresponding vertex pairs is outside the threshold range, the predetermined distance is changed until the number of selected corresponding vertex pairs falls within the threshold range.

  One skilled in the art will appreciate that the above-described method of performing deformable registration of a single object in an image or multiple objects in an image can be realized. When multiple objects are in the image, the vertices of all objects are combined into a corresponding cloud, one for each object. The method is then performed on each corresponding vertex group. Note also that the object being imaged need not have the same number of vertices. If the amount of vertices is not the same in the corresponding object, the correspondence can be defined based on the object with a smaller number of vertices. An object with a larger number of vertices will have extra vertices excluded, i.e. those that do not give rise to corresponding vertex pairs.

  The invention has been described with reference to one or more preferred embodiments. Obviously, others will come up with modifications and variations upon reading and understanding this specification. All such modifications and variations are intended to be included so long as they fall within the scope of the appended claims or their equivalents.

FIG. 5 shows the first imaged bladder as a triangular surface model with vertex subsets. FIG. 4 shows a second imaged bladder with a vertex subset as a triangular surface model. FIG. 5 shows a correspondence between first and second imaged bladders based on vertex subsets. FIG. 5 shows an exemplary method flow chart for calculating a volumetric deformation of an object to be imaged based on a selected subset of the object's triangular mesh vertices. FIG. 6 illustrates an example method for defining a selected subset of corresponding vertex pairs.

Claims (19)

In a method for selecting vertices to perform a deformable registration on an object to be imaged,
Determining the positions of a plurality of vertices in the first and second imaging objects;
Determining a set of corresponding vertex pairs, each having a vertex from a first imaging object and a vertex from a second imaging object;
Determining the distance between each corresponding vertex pair;
Sorting the set of corresponding vertex pairs based on the distance between each corresponding vertex pair;
Selecting the corresponding vertex pair having the maximum distance;
Excluding corresponding vertex pairs within a selected distance from the selected vertex pairs;
Selecting the corresponding vertex pair having the second largest distance and excluding corresponding vertex pairs that are within the selected distance from the newly selected vertex pair, all corresponding Repeating the selection step and the exclusion step until a vertex pair is either selected or excluded.   The method of claim 1, wherein the selected distance is a distance between vertices of the selected vertex pair.   The method of claim 1, wherein the selected distance is a predetermined value.   The method of claim 1, wherein the selected distance can be varied.   The method of claim 1, further comprising determining the number of selected corresponding vertex pairs.   6. The method of claim 5, further comprising modifying the selected distance if the number of selected vertex pairs is not within a predetermined threshold range.   The method of claim 1, wherein the number of vertices in the first imaging object is not equal to the number of vertices in the second imaging object.   The method of claim 1, wherein each of the first and second imaging objects comprises a plurality of anatomical objects.   The method of claim 1, wherein a plurality of vertices in the first and second imaging objects are obtained from a triangular surface mesh. An apparatus for selecting vertices to perform a deformable registration on an imaging object,
Means for determining the positions of a plurality of vertices in the first and second imaging objects;
Means for determining a set of corresponding vertex pairs, each having a vertex from a first imaging object and a vertex from a second imaging object;
Means for sorting the set of corresponding vertex pairs based on the distance between each corresponding vertex pair;
Means for selecting the corresponding vertex pair having the maximum distance;
Means for excluding corresponding vertex pairs that are within a selected distance from the selected vertex pair;
Means for selecting the corresponding vertex pair having the second largest distance and excluding corresponding vertex pairs within the selected distance from the newly selected vertex pair, all corresponding Means for repeating the selection and exclusion of the corresponding vertex pair until either the vertex pair is selected or excluded.   The apparatus of claim 10, wherein the selected distance is a distance between vertices of the selected vertex pair.   The apparatus of claim 10, wherein the selected distance is a predetermined value.   The apparatus of claim 10, wherein the selected distance can be varied.   11. Apparatus according to claim 10, comprising means for determining the number of selected corresponding vertex pairs.   15. The apparatus of claim 14, further comprising means for modifying the selected distance if the number of selected vertex pairs is not within a predetermined threshold range.   The apparatus of claim 10, wherein the number of vertices in the first imaging object is not equal to the number of vertices in the second imaging object.   The apparatus of claim 10, wherein each of the first and second imaging objects has a plurality of anatomical objects. The apparatus of claim 10 , wherein a plurality of vertices in the first and second imaging objects are obtained from a triangular surface mesh. In a method for performing a deformable registration on an object to be imaged,
Determining the positions of a plurality of vertices in the first and second imaging objects;
Determining a set of corresponding vertex pairs, each having a vertex from a first imaging object and a vertex from a second imaging object;
Determining the distance between each corresponding vertex pair;
Sorting the set of corresponding vertex pairs based on the distance between each corresponding vertex pair;
Selecting the corresponding vertex pair having the maximum distance;
Excluding corresponding vertex pairs within a selected distance from the selected vertex pairs;
Selecting the corresponding vertex pair having the second largest distance and excluding corresponding vertex pairs that are within the selected distance from the newly selected vertex pair, all corresponding Repeating the selection step and the exclusion step until a vertex pair is either selected or excluded;
Using the selected corresponding vertex pair to perform deformable registration on the imaging object.

Images